Department of Mathematics, University of Texas, Austin, TX 78712, USA

Transcription

1 PROFINITE PROPERTIES OF DISCRETE GROUPS ALAN W. REID Department of Mathematics, University of Texas, Austin, TX 78712, USA 1 Introduction This paper is based on a series of 4 lectures delivered at Groups St Andrews The main theme of the lectures was distinguishing finitely generated residually finite groups by their finite quotients. The purpose of this paper is to expand and develop the lectures. The paper is organized as follows. In 2 we collect some questions that motivated the lectures and this article, and in 3 discuss some examples related to these questions. In 4 we recall profinite groups, profinite completions and the formulation of the questions in the language of the profinite completion. In 5, we recall a particular case of the question of when groups have the same profinite completion, namely Grothendieck s question. In 6 we discuss how the methods of L 2 -cohomology can be brought to bear on the questions in 2, and in 7, we give a similar discussion using the methods of the cohomology of profinite groups. In 8 we discuss the questions in 2 in the context of groups arising naturally in low-dimensional topology and geometry, and in 9 discuss parafree groups. Finally in 10 we collect a list of open problems that may be of interest. Acknowledgement: The material in this paper is based largely on joint work with M. R. Bridson, and with M. R. Bridson and M. Conder and I would like to thank them for their collaborations. I would also like to thank the organizers of Groups St Andrews 2013 for their invitation to deliver the lectures, for their hopsitality at the conference, and for their patience whilst this article was completed. This work was supported in part by NSF grants. 2 The motivating questions We begin by recalling some terminology. A group Γ is said to be residually finite (resp., residually nilpotent, residually-p, residually torsion-free-nilpotent) if for each non-trivial γ Γ there exists a finite group (resp., nilpotent group, p-group, torsionfree-nilpotent group) Q and a homomorphism φ : Γ Q with φ(γ) If a finitely-generated group Γ is residually finite, then one can recover any finite portion of its Cayley graph by examining the finite quotients of the group. It is therefore natural to wonder whether, under reasonable hypotheses, the set might determine Γ up to isomorphism. C(Γ) = {G : G is a finite quotient of Γ}

2 Reid: Profinite properties of discrete groups 2 Assuming that the groups considered are residually finite is a natural condition to impose, since, first, this guarantees a rich supply of finite quotients, and secondly, one can always form the free product Γ S where S is a finitely generated infinite simple group, and then, clearly C(Γ) = C(Γ S). Henceforth, unless otherwise stated, all groups considered will be residually finite. The basic motivating question of this work is the following due to Remesselenikov: Question 1: If F n is the free group of rank n, and Γ is a finitely-generated, residually finite group, then does C(Γ) = C(F n ) imply that Γ = F n? This remains open at present, although in this paper we describe progress on this question, as well as providing structural results about such a group Γ (should it exist) as in Question 1. Following [31], we define the genus of a finitely generated residually finite group Γ to be: G(Γ) = { : C( ) = C(Γ)}. This definition is taken, by analogy with the theory of quadratic forms over Z where two integral quadratic forms can be locally equivalent (i.e., at all places of Q), but not globally equivalent over Z. Question 2: G(Γ) = {Γ}? Question 3: G(Γ) > 1? Question 4: groups? Which finitely generated (respectively, finitely presented) groups Γ have Which finitely generated (respectively, finitely presented) groups Γ have How large can G(Γ) be for finitely generated (resp., finitely presented) Question 5: What group theoretic properties are shared by (resp., are different for) groups in the same genus? In addition, if P is a class of groups, then we define G(Γ, P) = { P : C( ) = C(Γ)}, and can ask the same questions upon restricting to groups in P Rather than restricting the class of groups in a genus, we can ask to distinguish finitely generated groups by restricting the quotient groups considered. A particularly interesting case of this is the following. Note first that, a group Γ is residually nilpotent if and only if Γ n = 1, where Γ n, the n-th term of the lower central series of Γ, defined inductively by setting Γ 1 = Γ and defining Γ n+1 = [x, y] : x Γ n, y Γ. Two residually nilpotent groups Γ and Λ are said to have the same nilpotent genus if they have the same lower central series quotients; i.e., Γ/Γ c = Λ/Λc for all c 1. Residually nilpotent groups with the same nilpotent genus as a free group are termed parafree. In [10] Gilbert Baumslag surveyed the state of the art concerning groups of

3 Reid: Profinite properties of discrete groups 3 the same nilpotent genus with particular emphasis on the nature of parafree groups. We will discuss this in more detail in 9 below. 3 Some examples We begin with a series of examples where one can say something about Questions We first prove the following elementary result. Proposition 3.1 Let Γ be a finitely generated abelian group, then G(Γ) = {Γ}. Proof Suppose first that G(Γ) and is non-abelian. We may therefore find a commutator c = [a, b] that is non-trivial. Since is residually finite there is a homomorphism φ : Q, with Q finite and φ(c) 1. However, G(Γ) and so Q is abelian. Hence φ(c) = 1, a contradiction. Thus is abelian. We can assume that Γ = Z r T 1 and = Z s T 2, where T i (i = 1, 2) are finite abelian groups. It is easy to see that r = s, for if r > s say, we can choose a large prime p such that p does not divide T 1 T 2, and construct a finite quotient (Z/pZ) r that cannot be a quotient of. In addition if T 1 is not isomorphic to T 2, then some invariant factor appears in T 1 say, but not in T 2. One can then construct a finite abelian group that is a quotient of T 1 (and hence Γ 1 ) but not of Γ 2. Note that the proof of Proposition 3.1 also proves the following. Proposition 3.2 Let Γ be a finitely generated group, and suppose that G(Γ). Then Γ ab = ab. In particular b 1 (Γ) = b 1 ( ) Remarkably, moving only slightly beyond Z to groups that are virtually Z, the situation is dramatically different. The following result is due to Baumslag [9]. We include a sketch of the proof. Theorem 3.3 There exists non-isomorphic meta-cyclic groups Γ 1 and Γ 2 for which C(Γ 1 ) = C(Γ 2 ). Indeed, both of these groups are virtually Z and defined as extensions of a fixed finite cyclic group F by Z. Sketch Proof What Baumslag actually proves in [9] is the following, and this is what we sketch a proof of: ( ) Let F be a finite cyclic group with an automorphism of order n, where n is different from 1, 2, 3, 4 and 6. Then there are at least two nonisomorphic cyclic extensions of F, say Γ 1 and Γ 2 with C(Γ 1 ) = C(Γ 2 ). Recall that the automorphism group of a finite cyclic group of order m is an abelian group of order φ(m). So in ( ) we could take F to be a cyclic group of order 11, which has an automorphism of order 5.

4 Reid: Profinite properties of discrete groups 4 Now let F = a be a cyclic group of order m, and assume that it admits an automorphism α of order n as in ( ). Assume that α(a) = a r. Now some elementary number theory (using that φ(m) > 2 by assumption) shows that we can find an integer l such that (l, n) = 1, and (i) α l α, and (ii) α l α 1. Now define Γ 1 = a, b a m = 1, b 1 ab = a r to be the split extension of F induced by α and Γ 2 = a, c a m = 1, c 1 ac = a rl be the split extension of F induced by α l. The key claims to be established are that Γ 1 and Γ 2 are non-isomorphic, and that they have the same genus. That the groups are non-isomorphic can be checked directly as follows. If θ : Γ 1 Γ 2 is an isomorphism, then θ must map the set of elements of finite order in Γ 1 to those in Γ 2 ; that is to say θ preserves F, and so induces an automorphism of F. Thus θ(a) = a s where (s, m) = 1. Moreover since the quotients Γ i /F = Z for i = 1, 2, it follows that θ(b) = c ɛ a t where ɛ = ±1 and t is an integer. Now consider θ(a r ). When θ(b) = ca t we get: α(a s ) = a rs = θ(a r ) = θ(bab 1 ) = (ca t )a s (ca t ) 1 = α l (a s ), and it follows that α = α l. A similar argument holds when θ(b) = c 1 a t to show α 1 = α l, both of which are contradictions to (ii) above. We now discuss proving that the groups are in the same genus. Setting P = Γ 1 Z, Baumslag [9] shows that P is isomorphic to Γ 2 Z. That Γ 1 and Γ 2 have the same genus now follows from a result of Hirshon [34] (see also [9]) where it is shown that (see Theorem 9 of [34]): Proposition 3.4 Suppose that A and B are groups with A Z = B Z, then C(A) = C(B) The case of nilpotent groups more generally is well understood due to work of Pickel [52]. We will not discuss this in any detail, other than to say that, in [52] it is shown that for a finitely generated nilpotent group Γ, G(Γ) consists of a finite number of isomorphism classes of nilpotent groups, and moreover, examples where the genus can be made arbitrarily large are known (see for example [58] Chapter 11). Similar results are also known for polycyclic groups (see [29] and [58]) From the perspective of this article, more interesting examples where the genus has cardinality greater than 1 (although still finite) are given by examples of lattices in semi-simple Lie groups. We refer the reader to [4] and [5] for details but we will provide a sketch of some salient points. Let Γ be a lattice in a semi-simple Lie group, for example, in what follows we shall take Γ = SL(n, R k ) where R k denotes the ring of integers in a number field k. A natural, obvious class of finite quotients of Γ, are those of the form SL(n, R k /I) where I R k is an ideal. Let π I denote the reduction homomorphism Γ SL(n, R k /I), and Γ(I) the kernel. Note that by Strong Approximation for SL n (see [53] Chapter 7.4 for example) π I is surjective for all I. A congruence subgroup of Γ is any subgroup

5 Reid: Profinite properties of discrete groups 5 < Γ such that Γ(I) < for some I. A group Γ is said to have the Congruence Subgroup Property (henceforth abbreviated to CSP) if every subgroup of finite index is a congruence subgroup. Thus, if Γ has CSP, then C(Γ) is known precisely, and in effect, to determine C(Γ) is reduced to number theory. Expanding on this, since R k is a Dedekind domain, any ideal I factorizes into powers of prime ideals. If I = P a i i, then it is known that SL(n, R k /I) = SL(n, R k /P a i i ). Thus the finite groups that arise as quotients of SL(n, R k ) are determined by those of the form SL(n, R k /P a i i ). Hence we are reduced to understanding how a rational prime p behaves in the extension k/q. This idea, coupled with the work of Serre [59] which has shed considerable light on when Γ has CSP, allows construction of non-isomorphic lattices in the same genus. Example: Let k 1 = Q( ) and k 2 = Q( 8 48). Let Γ 1 = SL(n, R k1 ) and Γ 2 = SL(n, R k2 ) (n 3). Then Γ 1 and Γ 2 have CSP (by [59]), are non-isomorphic (by rigidity) and C(Γ 1 ) = C(Γ 2 ). The reason for the last statement is that the fields k 1 and k 2 are known to be adelically equivalent (see [36]); i.e. their Adele rings are isomorphic. This can be reformulated as saying that if V i (i = 1, 2) are the sets of valuations associated to the prime ideals in k 1 and k 2, then there is a bijection φ : V 1 V 2 such that for all ν V 1 we have isomorphisms (k 1 ) ν = (k2 ) φ(ν). This has, as a consequence, the desired identical splitting behavior of rational primes in k 1 and k Unlike in the previous subsections, there are recent examples of Bridson [14] of finitely presented groups Γ for which G(Γ) is infinite. This will be discussed further in Profinite methods An important reformulation of the discussion in 2 uses the language of profinite groups. In particular, the language of profinite completions is a particularly convenient formalism for organizing finite quotients of a discrete group. For completeness we provide some discussion of profinite groups and profinite completions of discrete groups. We refer the reader to [56] for a more detailed account of the topics covered here A directed set is a partially ordered set I such that for every i, j I there exists k I such that k i and k j. An inverse system is a family of sets {X i } {i I}, where I is a directed set, and a family of maps φ ij : X i X j whenever i j, such that: φ ii = id Xi ; φ ij φ jk = φ ik, whenever i j k. Denoting this system by (X i, φ ij, I), the inverse limit of the inverse system (X i, φ ij, I) is the set { lim X i = (x i ) } X i φ ij (x i ) = x j, whenever i j. i I

6 Reid: Profinite properties of discrete groups 6 We record the following standard facts about the inverse limit (see [56] Chapter 1 for further details): (i) Let (X i, φ ij, I) be an inverse system of non-empty compact, Hausdorff, totally disconnected topological spaces (resp. topological groups) over the directed set I, then lim X i is a non-empty, compact, Hausdorff, totally disconnected topological space (resp. topological group). (ii) Let (X i, φ ij, I) be an inverse system. A subset J I is defined to be cofinal, if for each i I, there exists j J with j i. If J is cofinal we may form an inverse system (X j, φ ij, J) obtained by omitting those i I \ J. The inverse limit lim X j can be identified with the image of lim X i under the projection map i I X i onto j J X j Returning to the world of group theory, let Γ be a finitely generated group (not necessarily residually finite for this discussion), and let N denote the collection of all finite index normal subgroups of Γ. Note that N is non-empty as Γ N, and we can make N into directed set by declaring that For M, N N, M N whenever M contains N. In this case, there are natural epimorphisms φ NM : Γ/N Γ/M, and the inverse limit of the inverse system (Γ/N, φ NM, N ) is denoted Γ and defined to be to the profinite completion of Γ. Note that there is a natural map ι : Γ Γ defined by g (gn) lim Γ/N, and it is easy to see that ι is injective if and only if Γ is residually finite. An alternative, perhaps more concrete way of viewing the profinite completion is as follows. If, for each N N, we equip each Γ/N with the discrete topology, then {Γ/N : N N } is a compact space and Γ can be identified with j(γ) where j : Γ {Γ/N : N N } is the map g (gn) From 4.1, Γ is a compact topological group, and so a subgroup U is open if and only if it is closed of finite index. In addition, a subgroup H < Γ is closed if and only if it is the intersection of all open subgroups of Γ containing it. More recently, it is a consequence of a deep theorem of Nikolov and Segal [50] that if Γ is a finitely generated group, then every finite index subgroup of Γ is open. Thus a consequence of this is the following elementary lemma (in which Hom(G, Q) denotes the set of homomorphisms from the group G to the group Q, and Epi(G, Q) denotes the set of epimorphisms). Lemma 4.1 Let Γ be a finitely-generated group and let ι : Γ Γ be the natural map to its profinite completion. Then, for every finite group Q, the map Hom( Γ, Q) Hom(Γ, Q) defined by g g ι is a bijection, and this restricts to a bijection Epi( Γ, Q) Epi(Γ, Q). We record the following corollary for later use.

7 Reid: Profinite properties of discrete groups 7 Corollary 4.2 If Γ 1 is finitely-generated and Γ 1 = Γ2, then for every finite group Q. Hom(Γ 1, Q) = Hom(Γ 2, Q) 4.4. The first Betti number of a finitely generated group is b 1 (Γ) = dim Q [(Γ/[Γ, Γ]) Z Q]. Given any prime p, one can detect b 1 (Γ) in the p-group quotients of Γ, since it is the greatest integer b such that Γ surjects (Z/p k Z) b for every k N. We exploit this observation as follows: Lemma 4.3 Let Λ and Γ be finitely generated groups. If Λ is isomorphic to a dense subgroup of Γ, then b 1 (Λ) b 1 (Γ). Proof For every finite group A, each epimorphism Γ A will restrict to an epimorphism on both Γ and Λ (since by density Λ cannot be contained in a proper closed subgroup). But the resulting map Epi( Γ, A) Epi(Λ, A) need not be surjective, in contrast to Lemma 4.1. Thus if Γ surjects (Z/p k Z) b then so does Λ (but perhaps not vice versa) We now discuss the profinite topology on the discrete group Γ, its subgroups and the correspondence between the subgroup structure of Γ and Γ. We begin by recalling the profinite topology on Γ. This is the topology on Γ in which a base for the open sets is the set of all cosets of normal subgroups of finite index in Γ. Now given a tower T of finite index normal subgroups of Γ: Γ > N 1 > N 2 >... > N k >... with N k = 1, this can be used to define an inverse system and thereby determines a completion of Γ T (in which Γ will inject). Now if the inverse system determined by T is cofinal (recall 4.1) then the natural homomorphism Γ Γ T is an isomorphism. That is to say T determines the full profinite topology of Γ. The following is important in connecting the discrete and profinite worlds (see [56] 3.2.2, where here we use [50] to replace open by finite index ). Notation of X in G. Given a subset X of a profinite group G, we write X to denote the closure Proposition 4.4 If Γ is a finitely generated residually finite group, then there is a one-to-one correspondence between the set X of subgroups of Γ that are open in the profinite topology on Γ, and the set Y of all finite index subgroups of Γ. Identifying Γ with its image in the completion, this correspondence is given by: For H X, H H.

8 Reid: Profinite properties of discrete groups 8 For Y Y, Y Y Γ. If H, K X and K < H then [H : K] = [H : K]. Moreover, K H if and only if K H, and H/K = H/K. The following corollary of this correspondence will be useful in what follows. Corollary 4.5 Let Γ be a finitely-generated group, and for each d N, let M d denote the intersection of all normal subgroups of index at most d in Γ. Then the closure M d of M d in Γ is the intersection of all normal subgroups of index at most d in Γ, and hence d N M d = 1. Proof If N 1 and N 2 are the kernels of epimorphisms from Γ to finite groups Q 1 and Q 2, then N 1 N 2 is the kernel of the extension of Γ Q 1 Q 2 to Γ, while N 1 N 2 is the kernel of the map Γ Q 1 Q 2 that one gets by extending each of Γ Q i and then taking the direct product. The uniqueness of extensions tells us that these maps coincide, and hence N 1 N 2 = N 1 N 2. The claims follow from repeated application of this observation. If now H < Γ, the profinite topology on Γ determines some pro topology on H and therefore some completion of H. To understand what happens in certain cases that will be of interest to us, we recall the following. Since we are assuming that Γ is residually finite, H injects into Γ and determines a subgroup H. Hence there is a natural epimorphism Ĥ H. This need not be injective. For this to be injective (i.e. the full profinite topology is induced on H) we require the following to hold: For every subgroup H 1 of finite index in H, there exists a finite index subgroup Γ 1 < Γ such that Γ 1 H < H 1. There are some important cases for which injectivity can be arranged. Suppose that Γ is a group and H a subgroup of Γ, then Γ is called H-separable if for every g G \ H, there is a subgroup K of finite index in Γ such that H K but g / K; equivalently, the intersection of all finite index subgroups in Γ containing H is precisely H. The group Γ is called LERF (or subgroup separable) if it is H-separable for every finitelygenerated subgroup H, or equivalently, if every finitely-generated subgroup is a closed subset in the profinite topology. It is important to note that even if the subgroup H of Γ is separable, it need not be the case that the profinite topology on Γ induces the full profinite topology on H. Stronger separability properties do suffice, however, as we now indicate. Lemma 4.6 Let Γ be a finitely-generated group, and H a finitely-generated subgroup of Γ. Suppose that Γ is H 1 -separable for every finite index subgroup H 1 in H. Then the profinite topology on Γ induces the full profinite topology on H; that is, the natural map Ĥ H is an isomorphism. Proof Since Γ is H 1 separable, the intersection of all subgroups of finite index in Γ containing H 1 is H 1 itself. From this it easily follows that there exists Γ 1 < Γ of finite index, so that Γ 1 H = H 1. The lemma follows from the discussion above.

9 Reid: Profinite properties of discrete groups 9 Subgroups of finite index obviously satisfy the conditions of Lemma 4.6, and if Γ is LERF, the conditions of Lemma 4.6 are also satisfied. Hence we deduce the following. Corollary 4.7 (1) If Γ is residually finite and H is a finite-index subgroup of Γ, then the natural map from Ĥ to H is an isomorphism. (2) If Γ is LERF and H is a finitely generated subgroup of Γ, then the natural map from Ĥ to H is an isomorphism. Another case of what the profinite topology does on a subgroup that will be of interest to us is the following. Let Γ be a residually finite group that is the fundamental group of a graph of groups. Let the edge groups be denoted by G e and the vertex groups by G v. The profinite topology on Γ is said to be efficient if it induces the full profinite topology on G v and G e for all vertex and edge groups, and G v and G e are closed in the profinite topology on Γ. The main example we will make use of is the following which is well-known: Lemma 4.8 Suppose that Γ is a free product of finitely many residually finite groups G 1,..., G n. Then the profinite topology on Γ is efficient. Proof Since Γ is residually finite, the trivial group is closed in the profinite topology. To see that each G i is closed in the profinite topology we prove that Γ is G i -separable. To that end let G denote one of the G i, and let g Γ \ G. Since g / G, the normal form for g contains at least one element a k G k G. Since G k is residually finite there is a finite quotient A of G k for which the image of a k is non-trivial. Using the projection homomorphism G 1... G n G k A defines a homomorphism for which the image of G is trivial but the image of g is not. This proves the vertex groups are closed. To see that the full profinite topology is induced on each G i, we need to show that for each G i, i = 1,..., n, the following condition holds (recall the condition for injectivity given above). For every subgroup H of finite index in G i, there exists a finite index subgroup H i < Γ such that H i G i < H. Let G denote one of the G i s and assume that H < G is a finite index subgroup. We can also assume that H is a normal subgroup. Then using the projection homomorphism Γ = G 1... G n G/H whose kernel K defines a finite index of subgroup of Γ with K G = H as required. Note that in the situation of Lemma 4.8, it also follows that Γ = Ĝ1 Ĝ2... Ĝn where indicates the profinite amalgamated product. We refer the reader to [56] Chapter 9 for more on this We now prove one of the key results that we make use of. This is basically proved in [25] (see also [56] pp ), the mild difference here, is that we employ [50] to replace topological isomorphism with isomorphism. Theorem 4.9 Suppose that Γ 1 and Γ 2 are finitely-generated abstract groups. Then Γ 1 and Γ 2 are isomorphic if and only if C(Γ 1 ) = C(Γ 2 ). Proof If Γ 1 and Γ 2 are isomorphic then the discussion following the correspondence provided by Proposition 4.4 shows that C(Γ 1 ) = C(Γ 2 ).

10 Reid: Profinite properties of discrete groups 10 For the converse, we argue as follows. For each n N let U n = {U : U is a normal subgroup of Γ 1 with [Γ 1 : U] n}, and V n = {V : V is a normal subgroup of Γ 2 with [Γ 2 : V ] n}. Then Γ 1 /U n C(Γ 1 ) and Γ 2 /V n C(Γ 2 ). Hence there exists a normal subgroup K < Γ 1 so that Γ 1 /K = Γ 2 /V n. Now it follows that K is an intersection of normal subgroups of index n, and so U n < K. Hence Γ 2 /V n = Γ 1 /K Γ 1 /U n. On reversing the roles of Γ 1 and Γ 2 reverses this inequality from which it follows that Γ 2 /V n = Γ1 /U n. Now for each such n, let A n denote the set of all isomorphisms Γ 1 /U n onto Γ 2 /V n. For each n this is a finite non-empty set with the property that for m n and α A n, then α induces a unique homomorphism f nm (α) : Γ 1 /U m Γ 2 /V m such that the following diagram commutes. Γ 1 /U n Γ 1 /U m α f nm (α) Γ 2 /V n Γ 2 /V m It follows that {A n, f nm } is an inverse system of (non-empty) finite sets, and so the inverse limit lim A n exists and defines an isomorphism of the inverse systems lim Γ 1/U n and lim Γ 2 /V n. Also note that since U n and V n are co-final, the discussion in 4.5 shows that they induce the full profinite topology on Γ 1 and Γ 2 respectively and so we have: Γ 1 = lim Γ 1 /U n = lim Γ 2 /V n = Γ2 as required. Thus statements about C(Γ) and G(Γ) can now be rephrased in terms of the profinite completion. For example, G(Γ) = { : = Γ} We now give some immediate applications of Theorem 4.9 and the previous discussion in the context of the motivating questions. Lemma 4.10 Let φ : Γ 1 Γ 2 be an epimorphism of finitely-generated groups. If Γ 1 is residually finite and Γ 1 = Γ2, then φ is an isomorphism. Proof Let k ker φ. If k were non-trivial, then since Γ 1 is residually finite, there would be a finite group Q and an epimorphism f : Γ 1 Q such that f(k) 1. This map f does not lie in the image of the injection Hom(Γ 2, Q) Hom(Γ 1, Q) defined by g g φ. Thus Hom(Γ 1, Q) > Hom(Γ 2, Q), contradicting Corollary 4.2. Definition 4.11 The rank d(γ) of a finitely-generated group Γ is the least integer k such that Γ has a generating set of cardinality k. The rank d(g) of a profinite group G is the least integer k for which there is a subset S G with k = S and S is dense in G.

11 Reid: Profinite properties of discrete groups 11 If Γ 1 is assumed to be a finitely generated free group of rank r and Γ 2 a finitely generated group with d(γ 2 ) = r and Γ 1 = Γ2, then it follows immediately from Lemma 4.10 that Γ 2 is isomorphic to a free group of rank r (using the natural epimorphism Γ 1 Γ 2 ). Indeed, one can refine this line of argument as follows. In the following proposition, we do not assume that Γ is residually finite. Proposition 4.12 Let Γ be a finitely-generated group and let F n be a free group. If Γ has a finite quotient Q such that d(γ) = d(q), and Γ = F n, then Γ = F n. Proof First Γ = F n, so Q is a quotient of F n. Hence n d(q). But d(q) = d(γ) and for every integer s d(γ) there exists an epimorphism F s Γ. Thus we obtain an epimorphism F n Γ, and application of the preceding lemma completes the proof. Corollary 4.13 Let Γ be a finitely-generated group. If Γ and its abelianisation have the same rank, then Γ = F n if and only if Γ = F n. Proof Every finitely-generated abelian group A has a finite quotient of rank d(a). As an application of Corollary 4.13 we give a quick proof that that free groups and surface groups are distinguished by their finite quotients. For if Γ is a genus g 1 surface group, then Γ and its abelianization have rank 2g. Corollary 4.13 then precludes such a group having the same profinite completion as a free group. Another application is the following. Another natural generalization of free groups are right angled Artin groups. Let K be a finite simplicial graph with vertex set V = {v 1,..., v n } and edge set E V V. Then the right angled Artin group (or RAAG) associated with K is the group A(K) given by the following presentation: A(K) = v 1,..., v n [v i, v j ] = 1 for all i, j such that {v i, v j } E. For example, if K is a graph with n vertices and no edges, then A(K) is the free group of rank n, while if K is the complete graph on n vertices, then A(K) is the free abelian group Z n of rank n. If the group Γ has a presentation of the form A R where A is finite and all of the relators r R lie in the commutator subgroup of the free group F (A), then both Γ and its abelianisation (which is free abelian) have rank A. The standard presentations of RAAGs have this form. Proposition 4.14 If Γ is a right-angled Artin group that is not free, then there exists no free group F such that F = Γ We shall also consider other pro-completions, and we briefly recall these. The pro-(finite nilpotent) completion, denoted Γ fn, is the inverse limit of the finite nilpotent quotients of Γ. Given a prime p, the the pro-p completion Γ p is the inverse limit of the finite p-group quotients of Γ. As above we have natural homomorphisms Γ Γ fn

12 Reid: Profinite properties of discrete groups 12 and Γ Γ p and these are injections if and only if Γ is residually nilpotent in the first case and residually p in the second. Note that in this language, two finitely generated residually nilpotent groups with the same nilpotent genus have isomorphic pro-(finite nilpotent) completions. This can proved in a similar manner as Proposition 4.4 using only the finite nilpotent quotients. Note that it is proved in [6] (before the general case of [50]) that for a finitely generated group Γ, every subgroup of finite index in Γ fn is open. Moreover, finitely generated groups in the same nilpotent genus also have isomorphic pro-p completions for all primes p. 5 Grothendieck Pairs and Grothendieck Rigidity A particular case of when discrete groups have isomorphic profinite completions is the following (which goes back to Grothendieck [28]) Let Γ be a residually finite group and let u : P Γ be the inclusion of a subgroup P. Then (Γ, P ) u is called a Grothendieck Pair if the induced homomorphism û : P Γ is an isomorphism but u is not. (When no confusion is likely to arise, it is usual to write (Γ, P ) rather than (Γ, P ) u.) Grothendieck [28] asked about the existence of such pairs of finitely presented groups and the first such pairs were constructed by Bridson and Grunewald in [15]. The analogous problem for finitely generated groups had been settled earlier by Platonov and Tavgen [54]. Both constructions rely on versions of the following result (cf. [54], [15] Theorem 5.2 and [13]). We remind the reader that the fibre product P < Γ Γ associated to an epimorphism of groups p : Γ Q is the subgroup P = {(x, y) : p(x) = p(y)}. Proposition 5.1 Let 1 N Γ Q 1 be a short exact sequence of groups with Γ finitely generated and let P be the associated fibre product. Suppose that Q 1 is finitely presented, has no proper subgroups of finite index, and H 2 (Q, Z) = 0. Then (1) (Γ Γ, P ) is a Grothendieck Pair; (2) if N is finitely generated then (Γ, N) is a Grothendieck Pair. More recently in [14], examples of Grothendieck Pairs were constructed so as to provide the first examples of finitely-presented, residually finite groups Γ that contain an infinite sequence of non-isomorphic finitely presented subgroups P n so that the inclusion maps u n : P n Γ induce isomorphisms of profinite completions. In particular, this provides examples of finitely presented groups Γ for which G(Γ) is infinite There are many classes of groups Γ that can never have a subgroup P for which (Γ, P ) is a Grothendieck Pair; as in [40], we call such groups Grothendieck Rigid. Before proving the next theorem, we make a trivial remark that is quite helpful. Suppose that H < Γ and Γ is H-separable, then (Γ, H) is not a Grothendieck Pair. The reason for this is that being separable implies that H is contained in (infinitely many) proper subgroups of Γ of finite index. In particular H < Γ is contained in proper subgroups of finite index in Γ. On the other hand if (Γ, H) is a Grothendieck Pair, H is dense in Γ and so cannot be contained in a closed subgroup (of finite index)

13 Reid: Profinite properties of discrete groups 13 of Γ. With this remark in place, we prove our next result. Recall that a group Γ is called residually free if for every non-trivial element g Γ there is a homomorphism φ g from Γ to a free group such that φ g (g) 1, and Γ is fully residually free if for every finite subset X Γ there is a homomorphism from Γ to a free group that restricts to an injection on X. Theorem 5.2 Let Γ be a finitely generated group isomorphic to either: a Fuchsian group, a Kleinian group, the fundamental group of a geometric 3-manifold, a fully residually free group. Then Γ is Grothendieck Rigid. Proof This follows immediately from the discussion above, and the fact that such groups are known to be LERF. For Fuchsian groups see [57], for Kleinian groups this follows from [2] and [62] and for fully residually free groups [60]. If M is a geometric 3-manifold, then the case when M is hyperbolic follows from the remark above, and when M is a Seifert fibered space see [57]. For those modelled on SOL geometry, separability of subgroups can be established directly and the result follows. Remark The case of finite co-volume Kleinian groups was proved in [40] without using the LERF assumption. Instead, character variety techniques were employed. In 8.2 we will establish Grothendieck Rigidity for prime 3-manifolds that are not geometric. 6 L 2 -Betti numbers and profinite completion Proposition 3.2 established that the first Betti number of a group is a profinite invariant. The goal of this section is to extend this to the first L 2 -Betti number, and to give some applications of this. We refer the reader to Lück s paper [47] for a comprehensive introduction to L 2 - Betti numbers. For our purposes, it is best to view these invariants not in terms of their original (more analytic) definition, but instead as asymptotic invariants of towers of finite-index subgroups. This is made possible by the Lück s Approximation Theorem [46]: Theorem 6.1 Let Γ be a finitely presented group, and let Γ = Γ 1 > Γ 2 >... > Γ m >... be a sequence of finite-index subgroups that are normal in Γ and intersect in the identity. Then for all p 0, the p-th L 2 -Betti number of Γ is given by the formula b (2) b p (Γ m ) p (Γ) = lim m [Γ : Γ m ]. An important point to note is that this limit does not depend on the tower, and hence is an invariant of Γ. We will mostly be interested in b (2) 1. Example 6.2 Let F be a free group of rank r. Euler characteristic tells us that a subgroup of index d in F is free of rank d(r 1) + 1, so by Lück s Theorem b (2) 1 (F r) = r 1. A similar calculation shows that if Σ is the fundamental group of a closed surface of genus g, then b (2) 1 (Σ) = 2g 2.

14 Reid: Profinite properties of discrete groups 14 Proposition 6.3 Let Λ and Γ be finitely presented residually finite groups and suppose that Λ is a dense subgroup of Γ. Then b (2) 1 (Γ) b(2) 1 (Λ). Proof For each positive integer d let M d be the intersection of all normal subgroups of index at most d in Γ, and let L d = Λ M d in Γ. We saw in Corollary 4.5 that d M d = 1, and so d L d = 1. Since Λ and Γ are both dense in Γ, the restriction of Γ Γ/M d to each of these subgroups is surjective, and hence [Λ : L d ] = [ Γ : M d ] = [Γ : M d ]. Now L d is dense in M d, while M d = M d, so Lemma 4.3 implies that b 1 (L d ) b 1 (M d ), and then we can use the towers (L d ) in Λ and (M d ) in Γ to compare L 2 -Betti numbers and find b (2) 1 (Γ) = lim d b 1 (M d ) [Γ : M d ] lim d b 1 (L d ) [Λ : L d ] = b(2) 1 (Λ), by Lück s approximation theorem. This has the following important consequence: Corollary 6.4 Let Γ 1 and Γ 2 be finitely-presented residually finite groups. If Γ1 = Γ 2, then b (2) 1 (Γ 1) = b (2) 1 (Γ 2). If one assumes only that the group Γ is finitely generated, then one does not know if the above limit exists, and when it does exist one does not know if it is independent of the chosen tower of subgroups. However, a weaker form of Lück s approximation theorem for b (2) 1 was established for finitely generated groups by Lück and Osin [48]. Theorem 6.5 If Γ is a finitely generated residually finite group and (N m ) is a sequence of finite-index normal subgroups with m N m = 1, then lim sup m b 1 (N m ) [Γ : N m ] b(2) 1 (Γ) We now give some applications of Proposition 6.3 in the context of Question 1 (and the analogous questions for Fuchsian groups). First we generalize the calculation in Example 6.2. Proposition 6.6 If Γ is a lattice in PSL(2, R) with rational Euler characteristic χ(γ), then b (2) 1 (Γ) = χ(γ). Proof It follows from Lück s approximation theorem that if H is a subgroup of index index d in Γ (which is finitely-presented) then b (2) 1 (H) = d b(2) 1 (Γ). Rational Euler characteristic is multiplicative in the same sense. Thus we may pass to a torsion-free subgroup of finite index in Γ, and assume that it is either a free group F r of rank r, or the fundamental group Σ g of a closed orientable surface of genus g. The free group case was dealt with above, and so we focus on the surface group case.

15 Reid: Profinite properties of discrete groups 15 Thus if Γ d is a subgroup of index d in Γ, then it is a surface group of genus d(g 1) + 1. The first Betti number in this case is 2d(g 1) + 1 and so b 1 (Γ d ) = 2 d χ(γ). Dividing by d = Γ : Γ d and taking the limit, we find b (2) 1 (Γ) = χ(γ). With this result and Proposition 6.3 we have the following. The only additional comment to make is that the assumption that the Fuchsian group Γ 1 is non-elementary implies it is not virtually abelian, and so b (2) 1 (Γ 1) 0. Corollary 6.7 Let Γ 1 be a finitely generated non-elementary Fuchsian group, and Γ 2 a finitely presented residually finite group with Γ 1 = Γ2. Then b (2) 1 (Γ 2) = b (2) 1 (Γ 1) = χ(γ 1 ) 0. Another standard result about free groups is that if F is a finitely generated free group of rank 2, then any finitely generated non-trivial normal subgroup of F has finite index (this also holds more generally for Fuchsian groups and limit groups, see [17] for the last statement). As a further corollary of Propositon 6.4 we prove the following. Corollary 6.8 Let Γ be a finitely presented residually finite group in the same genus as a finitely generated free group, and let N < Γ be a non-trivial normal subgroup. If N is finitely generated, then Γ/N is finite. Proof Proposition 3.1 shows that the genus of the infinite cyclic group contains only itself, and so we can assume that Γ lies in the genus of a non-abelian free group. Thus, by Corollary 6.4, b (2) 1 (Γ) 0. The proof is completed by making use of the following theorem of Gaboriau (see [27] Theorem 6.8): Theorem 6.9 Suppose that 1 N Γ Λ 1 is an exact sequence of groups where N and Λ are infinite. If b (2) 1 (N) <, then b (2) 1 (Γ) = 0. Indeed, using Theorem 6.5, Corollary 6.8 can be proved under the assumption that Γ is a finitely generated residually finite group. In this case, the argument establishes that if Γ is in the same genus as a finitely generated free group F, then b (2) 1 (Γ) b(2) 1 (F ) and we can still apply [27]. As remarked upon earlier, Question 1 is still unresolved, and in the light of this, Corollary 6.8 provides some information about the structural properties of a finitely generated group in the same genus as a free group. In 8.3, we point out some other properties that occur assuming that a group Γ is in the same genus as a finitely generated free group.

16 Reid: Profinite properties of discrete groups 16 Remark Unlike the case of surface groups, if M is a closed 3-manifold, then typically b (2) 1 (π 1(M)) = 0. More precisely, we have the following from [44]. Let M = M 1 #M 2 #... #M r be the connect sum of closed (connected) orientable prime 3-manifolds and that π 1 (M) is infinite. Then b (2) 1 (π 1(M)) = (r 1) r j=1 1 π 1 (M j ), where in the summation, if π 1 (M j ) is infinite, the term in the sum is understood to be zero Corollary 6.4 establishes that b (2) 1 is an invariant for finitely presented groups in the same genus. A natural question arises as to whether anything can be said about the higher L 2 -Betti numbers. Using the knowledge of L 2 -Betti numbers of locally symmetric spaces (see [21]), it follows that the examples given 3.4 will have all b (2) p equal. On the other hand, using [4] examples can be constructed which do not have all b (2) p being equal. Further details will appear elsewhere. 7 Goodness In this section we discuss how cohomology of profinite groups can be used to inform about Questions We begin by recalling the definition of the continuous cohomology of profinite groups (also known as Galois cohomology). We refer the reader to [59] and [56, Chapter 6] for details about the cohomology of profinite groups. Let G be a profinite group, M a discrete G-module (i.e., an abelian group M equipped with the discrete topology on which G acts continuously) and let C n (G, M) be the set of all continous maps G n M. One defines the coboundary operator d : C n (G, M) C n+1 (G, M) in the usual way thereby defining a complex C (G, M) whose cohomology groups H q (G; M) are called the continuous cohomology groups of G with coefficients in M. Note that H 0 (G; M) = {x M : gx = x g G} = M G is the subgroup of elements of M invariant under the action of G, H 1 (G; M) is the group of classes of continuous crossed homomorphisms of G into M and H 2 (G; M) is in one-to-one correspondence with the (equivalence classes of) extensions of M by G Now let Γ be a finitely generated group. Following Serre [59], we say that a group Γ is good if for all q 0 and for every finite Γ-module M, the homomorphism of cohomology groups H q ( Γ; M) H q (Γ; M) induced by the natural map Γ Γ is an isomorphism between the cohomology of Γ and the continuous cohomology of Γ. Example 7.1 Finitely generated free groups are good.

17 Reid: Profinite properties of discrete groups 17 To see this we argue as follows. As is pointed out by Serre ([59] p. 15), for any (finitely generated) discrete group Γ, one always has isomorphisms H q ( Γ; M) H q (Γ; M) for q = 0, 1. Briefly, using the description of H 0 given above (and the discrete setting), isomorphism for H 0 follows using denseness of Γ in Γ and discreteness of M. For H 1, this follows using the description of H 1 as crossed homomorphisms. If Γ is now a finitely generated free group, since H2 (Ĝ; M) is in one-to-one correspondence with the (equivalence classes of) extensions of M by Γ, it follows that H 2 ( Γ; M) = 0 (briefly, like the case of the discrete free group there are no interesting extensions). The higher cohomology groups H q ( Γ; M) (q 3) can also be checked to be zero. For example, since H q (Γ; M) = 0 for all q 2, the induced map H q ( Γ; M) H q (Γ; M) is surjective for all q 2, and it now follows from a lemma of Serre [59] (see Ex 1 Chapter 2, and also Lemma 2.1 of [41]) that H q ( Γ; M) H q (Γ; M) is injective for all q 2. We also refer the reader to the discussion below on cohomological dimension for another approach. Goodness is hard to establish in general. One can, however, establish goodness for a group Γ that is LERF if one has a well-controlled splitting of the group as a graph of groups [30]. In addition, a useful criterion for goodness is provided by the next lemma due to Serre (see [59, Chapter 1, Section 2.6]) Lemma 7.2 The group Γ is good if there is a short exact sequence 1 N Γ H 1, such that H and N are good, N is finitely-generated, and the cohomology group H q (N, M) is finite for every q and every finite Γ-module M. We summarize what we will need from this discussion. Theorem 7.3 The following classes of groups are good. Finitely generated Fuchsian groups. The fundamental groups of compact 3-manifolds. Fully residually free groups. Right angled Artin groups. Proof The first and third parts are proved in [30] using LERF and well-controlled splittings of the group, and the fourth is proved in [41]. The second was proved by Cavendish in his PhD thesis [23]. We will sketch the proof when M is closed. Note first that by [30] free products of residually finite good groups are good, so it suffices to establish goodness for prime 3-manifolds. As is shown in [30] goodness is preserved by commensurability, and so finite groups are clearly good. Thus it remains to establish goodness for prime 3-manifolds with infinite fundamental group. For geometric closed 3-manifolds, goodness will follow immediately from Lemma 7.2 (using the first part of the theorem) when Γ = π 1 (M) and M is a Seifert fibered space or has SOL geometry. For hyperbolic 3-manifolds the work of Agol [2] and Wise [62] shows that any finite volume hyperbolic 3-manifold has a finite cover that fibers over the circle, and once again by Lemma 7.2 (and the first part of the theorem) we deduce

18 Reid: Profinite properties of discrete groups 18 goodness. For manifolds with a non-trivial JSJ decomposition, goodness is proved in [61] Let G be a profinite group. Then the p-cohomological dimension of G is the least integer n such that for every finite (discrete) G-module M and for every q > n, the p-primary component of H q (G; M) is zero, and this is denoted by cd p (G). The cohomological dimension of G is defined as the supremum of cd p (G) over all primes p, and this is denoted by cd(g). We also retain the standard notation cd(γ) for the cohomological dimension (over Z) of a discrete group Γ. A basic connection between the discrete and profinite versions is given by Lemma 7.4 Let Γ be a discrete group that is good. If cd(γ) n, then cd( Γ) n. Proof If cd(γ) n then H q (Γ, M) = 0 for every Γ-module M and every q > n. By goodness this transfers to the profinite setting in the context of finite modules. Discrete groups of finite cohomological dimension (over Z) are torsion-free. In connection with goodness, we are interested in conditions that allow one to deduce that Γ is also torsion-free. For this we need the following result that mirrors the behavior of cohomological dimension for discrete groups (see [59, Chapter 1 3.3]). Proposition 7.5 Let p be a prime, let G be a profinite group, and H a closed subgroup of G. Then cd p (H) cd p (G). This quickly yields the following that we shall use later. Corollary 7.6 Suppose that Γ is a residually finite, good group of finite cohomological dimension over Z. Then ˆΓ is torsion-free. Proof If Γ were not torsion-free, then it would have an element x of prime order, say q. Since x is a closed subgroup of Γ, Proposition 7.5 tells us that cd p ( x ) cd p ( Γ) for all primes p. But H 2k ( x ; F q ) 0 for all k > 0, so cd q ( x ) and cd q ( Γ) are infinite. Since Γ is good and has finite cohomological dimension over Z, we obtain a contradiction from Lemma 7.4. Note that this can be used to exhibit linear groups that are not good. For example, in [45], it is shown that there are torsion-free subgroups Γ < SL(n, Z) (n 3) of finite index, for which Γ contains torsion of all possible orders. As a corollary of this we have: Corollary 7.7 For all n 3, any subgroup of SL(n, R) commensurable with SL(n, Z) is not good When the closed subgroup is a p-sylow subgroup G p (i.e., a maximal closed pro-p subgroup of G) then we have the following special case of Proposition 7.5 (see [56] 7.3). Note that cohomology theory of pro-p groups is easier to understand than general profinite groups, and so the lemma is quite helpful in connection with computing cohomology of profinite groups.

19 Reid: Profinite properties of discrete groups 19 Lemma 7.8 Let G p be a p-sylow subgroup of G. Then: cd p (G) = cd p (G p ) = cd(g p ). cd(g) = 0 if and only if G = 1. cd p (G) = 0 if and only G p = 1. Example 7.9 Let F be a finitely generated free group. Since a p-sylow subgroup of F is Z p, Lemma 7.8 gives an efficient way to establish that cd( F ) = In this subsection we point out how goodness (in fact a weaker property suffices) provides a remarkable condition to establish residual finiteness of extensions. First suppose that we have an extension: 1 N E Γ 1. Using right exactness of the profinite completion (see [56] Proposition 3.2.5), this short exact sequence always determines a sequence: N Ê Γ 1. To ensure that the induced homorphism N Ê is injective is simply again the statement that the full profinite topology is induced on N. As was noticed by Serre [59], this is guaranteed by goodness. Indeed the following is true, the proof of which we discuss below (the proof is sketched in [59] and see also [30] and [41]). Proposition 7.10 The following are equivalent for a group Γ. For any finite Γ-module M, the induced map H 2 ( Γ; M) H 2 (Γ; M) is an isomorphism; For every group extension 1 N E Γ 1 with N finitely generated, the map N Ê is injective. Before discussing this we deduce the following. Corollary 7.11 Suppose that Γ is residually finite and for any finite Γ-module M, the induced map H 2 ( Γ; M) H 2 (Γ; M) is an isomorphism. Then any extension E (as above) by a finitely generated residually finite group N is residually finite. Groups as in Corollary 7.11 are called highly residually finite in [41], and super residually finite in [22]. Proof By Proposition 7.10, and referring to the diagram below, we have exact sequences with vertical homomorphisms i N and i Γ being injective by residual finiteness. Now the squares commute, and so a 5-Lemma argument implies that i E is injective; i.e., E is residually finite. 1 N E Γ 1 ie iγ i N 1 N Ê Γ 1

20 Reid: Profinite properties of discrete groups 20 Proof We discuss the if direction below, and refer the reader to [41] for the only if. We will show that it suffices to prove the result with N finite. For then the case of N finite is dealt with by Proposition 6.1 of [30]. Thus assume that N is finitely generated and J a finite index subgroup of N. Recall that from 4.5 we need to show that there exists a finite index subgroup E 1 < E such that E 1 N < J. To that end, since N is finitely generated we can find a characteristic subgroup H < J of finite index in N that is normal in E. Thus we have: 1 N E Γ 1 π 1 N/H E/H Γ 1 Assuming that the result holds for the case of N finite we can appy this to N/H. That is to say we can find E 0 < E/H such that E 0 (N/H) = 1. Set E 0 = π 1 (E 0 ), then E 0 N < H < J as required. Given Corollary 7.11 and Theorem 7.3 we have: Corollary 7.12 Let Γ be a group as in Theorem 7.3. Then Γ is highly residually finite. Examples of groups that are not highly residually finite are SL(3, Z) (see [33]) and Sp(2g, Z) ([24]). In particular in [24] lattices in a connected Lie group are constructed that are not residually finite. These arise as extensions of Sp(2g, Z) We now return to Question 1, and in particular deduce some consequences about a group Γ in the same genus as a finitely generated free group. To that end, the following simple observation will prove useful. Corollary 7.13 Let Γ 1 and Γ 2 be finitely-generated (abstract) residually finite groups with Γ 1 = Γ2. Assume that Γ 1 is good and cd(γ 1 ) = n <. Furthermore, assume that H is a good subgroup of Γ 2 for which the natural mapping Ĥ Γ 2 is injective. Then H q (H; F p ) = 0 for all q > n. Proof If H q (H; F p ) were non-zero for some q > n, then by goodness we would have H q (Ĥ; F p) 0, so cd p (Ĥ) q > n. Now Ĥ Γ 2 is injective and so Ĥ = H. Hence Γ 1 contains a closed subgroup of p-cohomological dimension greater than n, a contradiction. Corollary 7.14 If Γ contains a surface group S, and Ŝ Γ is injective, then Γ is not isomorphic to the profinite completion of any free group. In particular, this also shows the following: Corollary 7.15 If L is a non-abelian free group, then L does not contain the profinite completion of any surface group, nor that of any free abelian group of rank greater than 1.